Power asymmetry reverses bot effects on cooperation in hybrid populations

The rise of artificial intelligence enables new ways to influence human cooperation with high controllability and broad scalability. While prior theoretical studies in symmetric snowdrift games suggest that cooperative bots can reduce cooperation and defective bots can enhance it, these conclusions rely on the assumption of equal payoffs across agents. Here, we extend this analysis to asymmetric human-machine hybrid populations, where normal players may receive higher or lower payoffs than bots depending on their relative power. We find that power asymmetry fundamentally reshapes the role of simple bots: when normal players are advantaged, cooperative bots suppress cooperation and defective bots enhance it; however, this effect reverses when normal players are disadvantaged, with cooperative bots promoting cooperation and defective bots undermining it. These findings hold across both structured and unstructured populations. Our results advance the understanding of how simple bots can be strategically used to influence cooperation and underscore the critical role of power asymmetry in hybrid systems.